Deep Learning for Improving Numerical Weather Prediction of Rainfall Extremes

Abstract

The accurate prediction of rainfall, and in particular rainfall extremes, remains challenging for numerical weather prediction models. This can be attributed to subgrid-scale parameterizations of processes that play a crucial role in the multi-scale dynamics, as well as the strongly intermittent nature and the highly skewed, non-Gaussian distribution of rainfall. Here we show that a specific type of deep neural networks can learn rainfall extremes from a numerical weather prediction ensemble. A frequency-based weighting of the loss function is proposed to enable the learning of extreme values in the distributions' tails. We apply our framework in a post-processing step to correct for errors in the model-predicted rainfall. Our method yields a much more accurate representation of relative rainfall frequencies and improves the forecast skill of extremes by factors ranging from two to above six, depending on the event magnitude.

A Systematic Review of Deep Learning Applications in Interpolation and Extrapolation of Precipitation Data

“…They showed that LSTMs significantly improved MAPLE forecasts for South Korea. By utilizing a loss function that penalizes rare extremes in rainfall data, Hess and Boers (2021) presented an ANN approach and showed that their approach notably improved rainfall forecasts over the TRMM product of NASA. For the Huaihe River basin in China, employed a CNN, as supplementary predictors to leverage geographical information and atmospheric circulation factors.…”